[PATCH] RISC-V: Clobber V registers on syscalls

[Rémi Denis-Courmont]

Le torstaina 15. kesäkuuta 2023, 23.33.44 EEST Palmer Dabbelt a écrit :
> > The kernel cannot simply clobber registers, as that would likely cause
> > data leakage from kernel to user mode. So it is unclear what the benefits
> > would be
> What's the data leakage?

Typically "clobbering" the register means that you are writing something else 
in them. If you don't restore them (or expressly reset them to zero or some 
other fixed value), then you leak daata.

Of course, if you don't actually use the register, then you don't leak 
anything in them. But then it's unclear what the benefit of marking them as 
clobbered is.

(...)
> It'd have to be a pretty special case: at least in libstdc++ and glibc
> the futex calls are behind function calls,

Traditionally, atomic variable methods are intrinsics, which result in either 
inline or outline C runtime calls (with some ad-hoc ABI that clobbers very 
little). They cannot be C functions, since they accept parameters of several 
different types.

atomic_notify_one, atomic_notify_all, and atomic_wait or however their 
standardised names end up, will presumably be outlines of the later type, that 
just happen to wrap futex() on Linux.

But anyway, if the spec says that registers are clobbered by system calls as 
Björn pointed out, then that's that.

-- 
Rémi Denis-Courmont
http://www.remlab.net/